dceluis / vacocam_render

Vision-Assisted Camera Orientation

Vision-Assisted Camera Orientation / VacoCam

Installation

git clone https://github.com/dceluis/vacocam_render
cd vacocam_render
pip install -r requirements.txt

Setup

A valid OpenAI API key is needed, if you want to use the AI supervision.

cp .env.sample .env

Then, add your key to .env

OPENAI_API_KEY="your_api_key"

Usage

Rendering a supervised video is a multi-step process.

First, we detect the balls on each frame:

python detect.py --model="./path/to/yolov8.pt" "./path/to/video.mp4"

We can now run the two-step supervision strategies.

1. Removing static clusters:

python track.py --tracking="declustered" "./path/to/video.mp4"

2. Supervise the video focus using gpt4-vision:

python track.py --tracking="vacocam" "./path/to/video.mp4"

Finally, we can render the video:

python render.py --max-zoom=1.9 --min-zoom=1.2 --max-area=400 --min-area=50 --vacocam "./path/to/video.mp4"

Technical Overview